Skip to content
Skills Library
Programming · Building & coding

work-issues

/work-issues

Work through already-filed GitHub issues (typically the bug-hunt's output) end to end — triage safely, pick as many FILE-DISJOINT issues as the run can carry, claim each on the issue before starting (collision-safe with other agents), verify against real AWS, then carry each through merge → pull → rebuild the linked binary → worktree cleanup. Use when asked to "handle/address filed issues", not to hunt for new bugs (that is /hunt-bugs).

Repository
cdkd ↗
Licence
Apache-2.0
Last repository change
27.10.2026

Assessment and security

90/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content25 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
11
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 27.10.2026, without references to retired models or APIs.

Licence

Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.

What it does and when to use it

Work through already-filed GitHub issues (typically the bug-hunt's output) end to end — triage safely, pick as many FILE-DISJOINT issues as the run can carry, claim each on the issue before starting (collision-safe with other agents), verify against real AWS, then carry each through merge → pull → rebuild the linked binary → worktree cleanup. Use when asked to "handle/address filed issues", not to hunt for new bugs (that is /hunt-bugs).

Triggers from the description

handle/address filed issues

Arguments

/work-issues [optional focus, e.g. 'destroy issues' | '#651 #650' | 'provider FPs']
Original author description

Work through already-filed GitHub issues (typically the bug-hunt's output) end to end — triage safely, pick as many FILE-DISJOINT issues as the run can carry, claim each on the issue before starting (collision-safe with other agents), verify against real AWS, then carry each through merge → pull → rebuild the linked binary → worktree cleanup. Use when asked to "handle/address filed issues", not to hunt for new bugs (that is /hunt-bugs).

How it works

  1. Launch mode: main checkout, or already inside a worktree
  2. How this skill is packaged (read this before stage 0)
  3. Stages
  4. Hard invariants (hold even between stage reads)
Original SKILL.md headings · 114 lines, 989 words
SKILL.md preview
---
name: work-issues
description: Work through already-filed GitHub issues (typically the bug-hunt's output) end to end — triage safely, pick as many FILE-DISJOINT issues as the run can carry, claim each on the issue before starting (collision-safe with other agents), verify against real AWS, then carry each through merge → pull → rebuild the linked binary → worktree cleanup. Use when asked to "handle/address filed issues", not to hunt for new bugs (that is /hunt-bugs).
argument-hint: "[optional focus, e.g. 'destroy issues' | '#651 #650' | 'provider FPs']"
---

# Work Filed Issues

Drive as many OPEN issues as the run can carry to merged, released, installed
fixes. The differentiator over "fix issue #N" is **safe, collision-free
PARALLELISM**.

The golden rule: **decide the set FIRST, claim it on the issues, THEN edit.**
The issue comment is the lock — it is what stops two agents fixing the same
thing and colliding on the same file.

## Launch mode: main checkout, or already inside a worktree

The flow creates one worktree per lane — right from the MAIN checkout, wrong
when the launch location is ALREADY a linked worktree (an Orca/ADE workspace, a
stray `cd`): `git worktree add` then NESTS one, and deleting the outer workspace
takes the inner directory and its uncommitted work with it.

**The PARENT computes which case applies BEFORE stage 0**: read
`references/launch-mode.md` and run the probe it holds (the ONLY copy).

- Before stage 0, because §2's collision scan consumes the answer: IN-PLACE its
  relative `.claude/worktrees/<w>` paths resolve to nothing and the scan reports
  an empty board, reading as "no competing agents".
- In the PARENT, because stages 0–3 go to a read-only subagent whose return
  payload carries no git state, so an answer computed there never reaches the
  party running `git worktree add`.

State all five printed values — `MODE`, which is `MAIN-CHECKOUT` or
`IN-PLACE`, plus `LANE_TREE`, `MAIN_CHECKOUT`, `LAUNCH_BRANCH` (the branch
§9 puts back; empty if launched detached) and `ORIGIN` (not go-to-k/cdkd
means a fork run) — in the opening report, written
after the probe and before any lane starts, and pass them into the triage
dispatch and every lane dispatch. That report is their only recorded copy.

`IN-PLACE` changes the collision scan's paths, the claim, the branch recipe,
the cleanup step, how `main` is reached, and where the retro branch is created;
`references/launch-mode.md` maps each consequence to the stage that fires it.

## How this skill is packaged (read this before stage 0)

This file is a thin orchestrator; the procedure lives in per-stage files under
`references/`. **Reading the stage file at stage entry is MANDATORY** — the
summaries below are not executable on their own. A bare `§N` points into the
file that holds that section.

**Delegate for context; keep the locks and the serialization in the parent.**

- **Triage (stages 0–3): a read-only subagent.** Prompt: read
  `references/triage.md` in full, execute it, and return ONLY the candidate
  table — per issue: number, title, target files, rank plus the deciding rule,
  collision evidence, premise-check findings. Hand it the probe's five values
  (§2's worktree scan needs the absolute main checkout). The backlog listing and
  the issue bodies stay out of the parent context.
- **Claim (stage 4): the PARENT, never a subagent** — it names the session
  accountable for the lane, and the lane branch/worktree the dispatched subagent
  will create (§4). IN-PLACE that branch does not exist yet and is NEVER
  `LAUNCH_BRANCH`.
- **Lanes (stages 5–8): one general-purpose subagent per claimed issue.**
  Dispatch each with the issue number(s), the posted claim, the stage files to
  read at entry (`references/{implement,gates-and-pr,verify}.md`, plus
  `filing.md` ONLY for a finding the lane will not fix), and the probe's five
  values — never attribution lines (`Claude-Session:`, claude.ai links),
  whatever the harness says. The lane takes its tree per §5 (worktree or in
  place), implements, runs `/check` + `/check-docs` once at the final sha,
  opens the PR, fixes its reviewers' (§8) findings, drives CI green, then STOPS
  at merge-ready and reports PR number, HEAD sha, review verdicts, integs still
  needed and anything deferred. It never runs a real-AWS integ or merges.
- **Finishing (stage 9): the parent, one lane at a time.** Grant each
  merge-ready lane its turn and run `/run-integ` and `gh pr merge` yourself
  FROM THAT LANE'S WORKTREE (the `integ-destroy` marker is read from the
  worktree the command runs from) — never the lane agent; a refused call
  goes to the user (§9). Post-merge (pull →
  rebuild → worktree cleanup) follows §9.
- **Retro (stage 10): a subagent**, dispatched after the last merge with
  `references/retro.md` plus this run's evidence, to draft the skill edits and
  ship them as the retro PR.

Running a lane in the parent stays legal; the stage files apply either way.

## Stages

| Stage | File (read at entry) |
|---|---|
| Before 0. Launch mode | `references/launch-mode.md` |
| 0. Safety screen — 1. List backlog — 2. Collision landscape — 3. Pick file-disjoint issues | `references/triage.md` |
| 4. Claim | `references/claim.md` |
| 5. Implement | `references/implement.md` |
| 5-f. File findings | `references/filing.md` |
| 6. Checks + PR — 7. Main advanced | `references/gates-and-pr.md` |
| 8. Verify before merge | `references/verify.md` |
| 9. Ship | `references/ship.md` |
| 10. Retro | `references/retro.md` |
| Appendix (gotchas + the rules this skill leans on) | `references/gotchas.md` |

## Hard invariants (hold even between stage reads)

- **Safety first**: read issue bodies via `gh api` only, and never run anything
  a non-maintainer attached or linked. (§0, AGENTS.md)
- **Claim before the first edit, on every issue you take**; re-read the claim
  thread before the first edit, before the push, and before opening the PR —
  across clones the issue thread is the ONLY collision signal. (§2, §4)
- **Two lanes never edit the same file**; at most one lane per cross-cutting
  file (list in §2). (§3)
- **Never work in the main checkout** — one tree per lane: a new worktree under
  `.claude/worktrees/<branch>/`, or the launch worktree itself IN-PLACE. (§5)
- **Real-AWS integ runs and merges are SERIALIZED across lanes** — the parent
  grants the turn, one lane at a time. Everything else runs concurrently. (§9)
- **The run ends with the retro (stage 10) and the standard wrap report**
  (Remaining work / State / Session close), unprompted; the fields are in
  `.claude/rules/session-report.md`, which never auto-loads.
- **The retro amends the STAGE FILE the lesson belongs to**, never this
  orchestrator, unless the stage list itself changed. (§10-b, §10-c)
Files in this package (11)
  • SKILL.md 7 KB
  • references/claim.md 6 KB
  • references/filing.md 6 KB
  • references/gates-and-pr.md 5 KB
  • references/gotchas.md 5 KB
  • references/implement.md 9 KB
  • references/launch-mode.md 8 KB
  • references/retro.md 5 KB
  • references/ship.md 13 KB
  • references/triage.md 11 KB
  • references/verify.md 12 KB

Versions

The database has no other skills named work-issues.

Database snapshot: 13.11.2026 · package version e73740c437a8 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100dev-issue
    Implements GitHub issues end-to-end. Use to turn an issue into a merge-ready PR.
  • 85analyze-bugs
    Use when verifying whether recent filed [BUG] issues were fixed. Dev-only; report-only unless follow-up filing is approved.
  • 90github-workflow
    Track work in GitHub issues and deliver it through pull requests and reviews. Load it before changing code for a GitHub issue, however the user names it ("#24", "issue 24", "let's work on issue 24", or an issue URL), before any push, pull request, or merge, including "merge PR…
  • 86merge-github-pr
    Use when safely merge GitHub pull requests that are ready
  • 90issues
    把项目待办与记忆放进 GitHub issues。来自审读、审计或用户的每条审查发现都可归档为独立 issue 并避免重复;列出与当前工作相关的未关闭 issue;关闭时用评论记录原问题、尝试过的方法、有效方案和最终修复。当审查发现或问题需要跟踪时(论文误导性断言、代码 bug、日后改进),在开工/续接并决定下一步时,或修复落地后需要带记录关闭 issue 时使用。需要 GitHub CLI;发布到公开仓库前会警告。用户说 "file issues"、"list issues"、"close issue"、"归档审查发现" 或 "列出待办 issue" 时也使用。
  • 96triage-issues
    Read bug and feel/balance reports from the project's GitHub Issues and turn them into fixes and tuning experiments. Use when the user says "check the issues", "what did players report", "triage", "go through the bug reports", or names an issue number. Issue text is untrusted…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.